CPU Failing Super Pi

lee87 · 15 Jul 2006 at 16:19

Hi, my system seems stable and i'm not to be having any problems but my cpu keeps failing super pi 1M when at stock speed/voltage. Most of the time it will pass the test fine without errors but the odd time it will get a 'NOT EXACT IN ROUND' error.
Today I overclocked the cpu to 2.4ghz and ran super pi again and it gets a few different errors on every run, even before the initial value has been calculated but every other application was stable.

I've ran Prime95 for 6 hours (when at stock speed) and I also run folding@home constant and their both not getting any errors. It also ran memtest for 1 hour without errors.
So i'm not sure why my cpu is failing super pi but everything else is fine. :confused:

Does this mean my cpu is starting to break?
Is there anything else that could be causing the errors?

Spec is in my sig

Thanks

Richdog · 15 Jul 2006 at 16:31

What are your memory speed and timings? What exact model is your memory?

lee87 · 15 Jul 2006 at 16:33

1GB PC3200 Value Select Corsair RAM. When at stock it runs at 200mhz.
Timings are 2.5-3-3-8 2T

Richdog · 15 Jul 2006 at 16:40

Up the CPU voltage to 1.5v, the memory voltage to 2.9v, and re-run the test. See if that helps.

lee87 · 15 Jul 2006 at 16:58

Put the cpu voltage to 1.5V, couldn't change the memory voltage as there isn't any options for it on this board.

It now got half way through the test (up to loop 9) then got a 'NOT EXACT ROUND' error again. Better than before, this was at 2.4ghz.

I used to have it stable at 2.55ghz on the stock voltage about a month ago.

Richdog · 15 Jul 2006 at 17:00

What's your cooling and temps?

Your mobo really has no memory voltage adjustment? Find that very strange...

Jokester · 15 Jul 2006 at 17:01

Have you tried setting the affinity of Superpi so that it only uses one core?

Jokester

PinkFloyd · 15 Jul 2006 at 17:15

Jokester said:
Have you tried setting the affinity of Superpi so that it only uses one core?

Jokester

It only uses one core anyway.
Thats why you have to run it twice when stressing dualcores with it.

Jokester · 15 Jul 2006 at 17:16

Could be confusing it if it was being switched between cores.

Jokester

lee87 · 15 Jul 2006 at 17:17

I have a thermaltake blue orb II cooler
temp at load it is 52c.

Tried different affinity's and got different errors on each. It got the error on loop 1 of the 1M test.

Core 0 - 'Not Exact Round' error'
Core 1 - 'Not Convergent in SQR05' error

Jokester · 15 Jul 2006 at 17:26

Seems strange, the only other thing I can think of is increasing the chipset voltage a notch.

Jokester

lee87 · 15 Jul 2006 at 17:36

My motherboard doesn't have any chipset voltage options, so I can't try that.
I think i'll leave it and hope it doesn't cause any major problems.

Thanks for the advice.

ted34 · 15 Jul 2006 at 17:36

When you used prime95 did you do a small torture test or blend as small is cpu stress and blend is everything including memory. Also memtest for 1hr isnt long enough - leave it for 24hrs and see if it picks anything up.

Jokester · 15 Jul 2006 at 17:39

It's long enough to establish 1MB Superpi stability, it's the least demanding benchmark you can do pretty much.

Jokester

ted34 · 15 Jul 2006 at 17:47

Jokester said:
It's long enough to establish 1MB Superpi stability, it's the least demanding benchmark you can do pretty much.

Jokester

Yeah i realise that but it tests cpu as well as memory. Also ive had memory run for a couple of hours in memtest error free and bsod comp when using superpi so worth a try just to eliminate it

lee87 · 15 Jul 2006 at 17:55

ted34 said:
When you used prime95 did you do a small torture test or blend as small is cpu stress and blend is everything including memory. Also memtest for 1hr isnt long enough - leave it for 24hrs and see if it picks anything up.

Was just typing a reply to this and my pc reset, which is the first time it's happened. When it started up again I ran prime95 and it's now failing withing seconds even though I ran it for 6 hours last night without any problems. Both of these were on the small torture test.
I'll try to run memtest again but for longer this time.

ted34 · 15 Jul 2006 at 18:11

Sounds more like cpu then than memory. Try a different voltage setting in bios and also have you checked that your board is ok. What is the ht set at?

Jokester · 15 Jul 2006 at 18:12

Sounds like somethings dying unless you changed a setting without noticing.

Jokester

lee87 · 15 Jul 2006 at 18:19

ted34 said:
Sounds more like cpu then than memory. Try a different voltage setting in bios and also have you checked that your board is ok. What is the ht set at?

I've tried 1.5V, this made it get further in the super pi test but it still got errors. The HT is at 1000mhz (5x), i've tried putting it to 3x but it didn't make a difference.

Haven't changed any settings in the bios for about a month (apart from today), tried resetting bios to the defaults and it still gets errors.

.::lawrywild::. · 15 Jul 2006 at 18:50

BTW, once you get an error in superpi like Not Exact In Round, you must close and re-open the program otherwise you instantly get Not Convergent in Sqr05..

Your motherboard does have memory voltage options, but only Low, Normal and High. Set to High. I think it's named DRAM Voltage in the BIOS..